The Interval Question

Your data is fine.
Your timing is not.

Your systems record constantly. You look monthly. Everything that happens between those two rates is invisible, and not because anything is broken.

The Inherited Interval

Nobody Chose Your
Reporting Interval

How often do you actually look at your numbers?

Not the dashboard you could open. The moment you sit down, look at a figure, and form a judgement about it. For most businesses that is monthly, in a meeting, about a period that has already closed.

That interval was inherited rather than chosen. It was set in an era when a person had to do the looking, and it was set by what that person could sustain and what the business could afford to pay them for.

It was never derived from the thing being measured.

I spent years doing exactly this by hand, on real systems, and I was too slow.

Ask YourselfWhen did you last look at a number and form a judgement about it? Not read it. Form a judgement about it, and decide whether to act.
The Sampling Problem

Under-Sampled Data
Does Not Go Quiet

There is a piece of mathematics that tells you whether your interval is good enough. It has been settled since 1928.

You have seen it working, in an old Western. The stagecoach speeds up, and the wheels appear to slow, stop, and then turn backwards.

The wheel is not turning backwards. The camera samples twenty four times a second, and the wheel is turning faster than that.

Notice what you actually see on screen. Not a blur. Not something obviously broken. A clear, steady, entirely believable picture of a wheel going the wrong way.

Harry Nyquist proved that to observe something properly you must sample it at least twice per cycle. Below that rate you do not lose a little detail. You get a confident, stable, plausible picture of something that is not happening.

"Under-sampled data does not go quiet. It lies to you fluently."

Actually happening What you see

The same data, looked at too slowly. Nothing here is inaccurate. Every reading is correct. The trend they appear to form does not exist.

Where The Value Goes

Recorded Every Minute.
Reviewed Once A Month.

Your systems already record constantly, in most cases every few minutes. Your looking happens monthly. Everything in the space between those two rates is invisible, and it is not invisible because the data is bad. It is invisible because of when you looked.

Be fair to yourself about why that gap exists, because it is not negligence. It is economics. Paying people to look often enough was never justifiable. Paying them to then interpret what they found, and decide what to do about it, was less justifiable still. So every business settled on a cadence that was affordable rather than a cadence that was sufficient.

That was the correct decision, for as long as a person had to do the looking.

That constraint moved about two years ago. The reporting cycle it created did not.

The PointThe cadence you run was affordable. It was never shown to be sufficient.
A Worked Example

A Sampling Problem
Wearing Solar Clothing

Solar is where this was tested properly, on live systems, before any of it was said out loud.

An AI observer sits inside every site on a monitoring platform now running across six installations and five manufacturers. It watches continuously. It builds a memory of how that specific installation behaves across hours, days, seasons and years. And it writes a plain sentence connecting what the system is doing to what it was bought to do.

One of those sites has a ten kilowatt hour battery. The battery is healthy. It is charged. The system is buying power from the grid anyway.

Nothing is broken. No alarm fired. Every dashboard would show that site as fine, because by every measure a dashboard understands, it is fine.

Now consider what the industry's own reliability body concedes. The International Energy Agency's photovoltaic programme classifies any performance loss below roughly two to three percent as below detection limit. Not too small to matter. Too small to see.

Underneath that threshold sits dust on glass. Dust alone removes somewhere between four and seven percent of the world's solar energy every year, worth billions of euros. Nobody decided to lose that money. It sits below the detection limit of the way the industry looks.

That is not a solar problem. It is a sampling problem wearing solar clothing.

Below Detection LimitThe industry standard for monitoring a solar plant cannot see a loss of two percent. Not will not. Cannot.
What Hides UnderneathDust on glass removes four to seven percent of the world's solar energy annually. Nobody decided to lose it. It sits below the threshold of the way we look.
The General Case

The Same Gap,
In Every Business

The constraint is completely general. Anywhere data accumulates faster than a person can review it, the same gap opens.

One

Process and Inspection

A process drifts continuously. The inspection interval was set at commissioning, not derived from how fast the drift actually moves.

Two

Debtors and Payment Behaviour

Payment behaviour changes within weeks. The ageing report arrives monthly, after the exposure has already been taken.

Three

Skills and Certification

Registrations and competencies lapse on their own schedule. Nobody looks until an audit does.

Four

Stock and Working Capital

Stock turns daily. The count that tells you what you actually hold happens once a quarter.

It is the same mathematics, the same economics, and the same gap.

What This Changes

From Report
To Conversation

The report is a strange artefact when you look at it directly.

The Report

Describes a period that has already closed.

Read when convenient, not when relevant.

Asks the reader to work out what matters.

One depth of analysis, fixed in advance by whoever built it.

The Conversation

Any period, asked about right now.

Analysis already done, at the right depth.

Stays current for as long as you talk.

Follow the thread as far as you have time for.

The channel matters. So does whether there is anything worth talking to on the other end.

The Question To Take Home

Nyquist Gives You The Rule.
It Does Not Give You Your Number.

How quickly does something go wrong in your business before it starts costing you money?

Not how often you report. How fast the problem actually moves.

Most people have never had to ask, because until recently there was no point in asking. Looking that often was not affordable. That constraint has gone. The reporting cycle it created has not.

If you do not know your number, every reporting interval you run is a guess.

It may well be a good guess. Nobody has checked.

On Automating The QuestionAsk a vague question, get a vague answer, then blame the AI. Somebody has to understand the system well enough to ask the question properly. That part is engineering, and it does not automate.

Classical sampling theory assumes a continuous band-limited signal, so applying it to business reporting is a motivated analogy rather than a theorem. It is a good analogy, and the arithmetic underneath it is real.

Where To Go From Here

Your Number,
Or The Worked Example

If You Want Your Own NumberA Monitoring Calibration examines what you measure, how often anyone actually looks, and how fast each thing can move. The finding is the gap between those three. It is a written document, presented in person, and it stands on its own.
Read About Monitoring Calibration
If You Want To Understand The MethodThe solar monitoring platform is where this was built and proven. Read how the observation works, what it detects that a dashboard cannot, and how a system is held against what it was bought to do.
Platform Page In Preparation

If neither is quite it, the shortest route is to describe what is happening and let the diagnosis follow.

Start a Conversation